It's possible for `IDiaSymbol::get_name` to return S_OK and provide
and empty string. I haven't figured out the exact root cause yet
(the symbols in question are coming from the Rust standard library),
but FUNC lines with missing function names break the processor and
so we should never do it. This change makes it output "<name omitted>"
which matches the behavior of the DWARF dumping code.
R=mark@chromium.org
BUG=https://bugzilla.mozilla.org/show_bug.cgi?id=1272278
Review URL: https://codereview.chromium.org/1985643004 .
This added debug fission support.
It tries to find the dwp file from the debug dir /usr/lib/debug/*/debug
and read symbols from them.
Most of this patch comes from
https://critique.corp.google.com/#review/52048295
and some fixes after that.
The elf_reader.cc comes from TOT google code. I just
removed some google dependency.
Current problems from this patch
1: Some type mismatch: from uint8_t * to char *.
2: Some hack to find the .dwp file. (replace .debug with .dwp)
BUG=chromium:604440
R=dehao@google.com, ivanpe@chromium.org
Review URL: https://codereview.chromium.org/1884283002 .
In Android, the mmap could be overlapped by /dev/ashmem, we adjusted
the range in https://breakpad.appspot.com/9744002/, but adjusted
range isn't written back to module, this caused the corresponding
module be dropped in BasicCodeModules copy constructor.
This also fix a lot of 'unable to store module' warnings
when dumping Android's minidump.
BUG=606972
R=mark@chromium.org, wfh@chromium.org
Review URL: https://codereview.chromium.org/1939333002 .
Patch from Tao Bai <michaelbai@chromium.org>.
Reason for revert:
It is causing breakpad crash reports to be invalid (see the associated
bug).
Merging empty holes in r-x mappings was originally introduced in
https://breakpad.appspot.com/7714003 to deal with the first generation
of relro packing, which could introduce holes within a .so mapping:
[libchrome.so]
[guard region]
[libchrome.so]
However, the logic is broken for the case of two *different* adjacent
.so mappings with a guard region in the middle:
[libfoo.so]
[guard region]
[libchrome.so]
In this case the guard region is mistakenly associated with libfoo.so,
but that is not the right thing to do. In fact, the second generation of
rerlo packing added the guard region to prevent mmaps from overlapping
and to give room for the non-zero vaddr of relro-packed libraries, which
require an anticipated load bias.
As the first generation of relro packing is not used anymore, there is
no reason to keep this buggy code, which causes failures in decoding
crashes where an arbitrary library is mapped immediately before a rerlo
packed library.
Original issue's description:
> Extend mapping merge to include reserved but unused mappings.
>
> When parsing /proc/pid/maps, current code merges adjacent entries that
> refer to the same library and where the start of the second is equal to
> the end of the first, for example:
>
> 40022000-40025000 r-xp 00000000 b3:11 827 /system/lib/liblog.so
> 40025000-40026000 r--p 00002000 b3:11 827 /system/lib/liblog.so
> 40026000-40027000 rw-p 00003000 b3:11 827 /system/lib/liblog.so
>
> When the system linker loads a library it first reserves all the address
> space required, from the smallest start to the largest end address, using
> an anonymous mapping, and then maps loaded segments inside that reservation.
> If the loaded segments do not fully occupy the reservation this leaves
> gaps, and these gaps prevent merges that should occur from occurring:
>
> 40417000-4044a000 r-xp 00000000 b3:11 820 /system/lib/libjpeg.so
> > 4044a000-4044b000 ---p 00000000 00:00 0
> 4044b000-4044c000 r--p 00033000 b3:11 820 /system/lib/libjpeg.so
> 4044c000-4044d000 rw-p 00034000 b3:11 820 /system/lib/libjpeg.so
>
> Where the segments that follow this gap do not contain executable code
> the failure to merge does not affect breakpad operation. However, where
> they do then the merge needs to occur. Packing relocations in a large
> library splits the executable segment into two, resulting in:
>
> 73b0c000-73b21000 r-xp 00000000 b3:19 786460
> /data/.../libchrome.2160.0.so
> > 73b21000-73d12000 ---p 00000000 00:00 0
> 73d12000-75a90000 r-xp 00014000 b3:19 786460
> /data/.../libchrome.2160.0.so
> 75a90000-75c0d000 rw-p 01d91000 b3:19 786460
> /data/.../libchrome.2160.0.so
>
> Here the mapping at 73d12000-75a90000 must be merged into 73b0c000-73b21000
> so that breakpad correctly calculates the base address for text.
>
> This change enables the full merge by also merging anonymous maps which
> result from unused reservation, identified as '---p' with offset 0, and
> which follow on from an executable mapping, into that executable mapping.
>
> BUG=chromium:394703
BUG=chromium:499747
R=primiano@chromium.org, rmcilroy@chromium.org
Review URL: https://codereview.chromium.org/1923383002 .
The x86-64 frame pointer-based unwind method will accept values
that aren't valid for the frame pointer register and the return address.
This fixes it to reject non-8-byte-aligned frame pointers, as
well as non-canonical addresses for the return address it finds.
A colleague of mine asked me why Breakpad gave a bad stack
for a crash in our crash-stats system:
https://crash-stats.mozilla.com/report/index/a472c842-2c7b-4ca7-a267-478cf2160405
Digging in, it turns out that the function in frame 0 is a leaf function,
so MSVC doesn't generate an entry in the unwind table for it, so
dump_syms doesn't produce a STACK CFI entry for it in the symbol file.
The stackwalker tries frame pointer unwinding, and %rbp is set to a
value that sort-of works, so it produces a garbage frame 1 and then
is lost. Either of the two checks in this patch would have stopped
the stackwalker from using the frame pointer.
It's possible we could do something smarter on the dump_syms side,
like enumerating all functions and outputing some default STACK CFI rule
for those that don't have unwind info, but that wouldn't fix crashes
from existing builds without re-dumping symbols for them. In any event,
these checks should always pass for valid frame pointer-using functions.
R=mark@chromium.org
BUG=https://bugzilla.mozilla.org/show_bug.cgi?id=1263001
Review URL: https://codereview.chromium.org/1902783002 .
Currently an inlined function in a namespace in DWARF will
be given a name comprised of just `namespace::`. This is due
to a logic error in ComputeQualifiedName, where it doesn't
handle an empty `unqualified_name` properly.
We apparently have a fair number of these in our Mac builds,
an example of the DWARF that's being mishandled looks like:
0x117eda40: TAG_namespace [5] *
AT_name( "js" )
AT_decl_file( "../../dist/include/js/Utility.h" )
AT_decl_line( 35 )
0x11808500: TAG_subprogram [251] *
AT_low_pc( 0x0000000002f12110 )
AT_high_pc( 0x0000000002f1216b )
AT_APPLE_omit_frame_ptr( 0x01 )
AT_frame_base( rsp )
AT_abstract_origin( {0x0000000011800a4f}"_ZN2js40TraceManuallyBarrieredGenericPointerEdgeEP8JSTracerPPNS_2gc4CellEPKc" )
AT_MIPS_linkage_name( "_ZN2js40TraceManuallyBarrieredGenericPointerEdgeEP8JSTracerPPNS_2gc4CellEPKc" )
AT_name( "TraceManuallyBarrieredGenericPointerEdge" )
AT_decl_file( "/builds/slave/rel-m-rel-m64_bld-000000000000/build/js/src/gc/Marking.cpp" )
AT_decl_line( 547 )
AT_external( 0x01 )
AT_APPLE_optimized( 0x01 )
AT_inline( DW_INL_inlined )
This turned a few instances of this in the file I was testing on into
`<name omitted>`, which seems to just be a symptom of the
"DW_AT_abstract_origin comes later in the file" issue. (Which is probably
also worth fixing given that it occurs some 29k times when dumping
symbols from Firefox's XUL binary, but it's a separate issue.)
R=mark@chromium.org
BUG=
Review URL: https://codereview.chromium.org/1887033002 .
I ran minidump_dump on a dump from Firefox on my Windows 10 machine
and noticed some streams that Breakpad didn't have names for.
Looking in minidumpapiset.h in the Windows 10 SDK finds these values
in MINIDUMP_STREAM_TYPE. There are also struct definitions for the
stream data for some of them (all but JavaScriptData), but I don't have
a particular need for those currently.
R=mark@chromium.org
BUG=
Review URL: https://codereview.chromium.org/1884943002 .
Doing a `make -jN check` from a fresh build breaks (and has probably been
broken for a while). linux_client_unittest_shlib is missing $(TEST_LIBS)
from its _DEPENDENCIES. The automake manual says if _DEPENDENCIES are not
specified they'll be computed from _LDADD, but we are specifying it and just
leaving out $(TEST_LIBS).
R=vapier@chromium.org
BUG=
Review URL: https://codereview.chromium.org/1870733005 .
GCC will still warn about unused return value with the form:
if (write(...));
Instead, change the semi-colon to an empty set of braces.
BUG=chromium:428478
TEST=build+test still works
It is often helpful to check if a particular symbol file dumped by
dump_syms actually matches a version of a binary file we have. The
symbol output contains an ID which can be used to see if it matches
the binary file. Unfortunately, this ID is internally calculated
and not a standard hash of the binary file. Being able to output the
header information only will allow users to determine whether their
symbol file is up to date or not.
R=jochen@chromium.org
BUG=561447
Review URL: https://codereview.chromium.org/1864823002 .
Patch from David Yen <dyen@chromium.org>.
Some projects will get build break because the comipler is confused when
searches for the standard stdio.h. Rename the wrapper file to avoid that.
renamed: src/common/stdio.h -> src/common/stdio_wrapper.h
modified: src/processor/minidump.cc
modified: src/processor/dump_context.cc
modified: src/processor/logging.cc
modified: src/processor/minidump.cc
modified: src/processor/minidump_processor.cc
modified: src/processor/stackwalk_common.cc
modified: src/processor/symbolic_constants_win.cc
R=mark@chromium.org, labath@google.com
Review URL: https://codereview.chromium.org/1864603002 .
Patch from Yunxiao Ma <yxma@google.com>.
This preserves full build ids in minidumps, which are useful for
tracking down the right version of system libraries from Linux
distributions.
The default build id produced by GNU binutils' ld is a 160-bit SHA-1
hash of some parts of the binary, which is exactly 20 bytes:
https://sourceware.org/binutils/docs-2.26/ld/Options.html#index-g_t_002d_002dbuild_002did-292
The bulk of the changes here are to change the signatures of the
FileID methods to use a wasteful_vector instead of raw pointers, since
build ids can be of arbitrary length.
The previous change that added support for this in the processor code
preserved the return value of `Minidump::debug_identifier()` as the
current `GUID+age` treatment for backwards-compatibility, and exposed
the full build id from `Minidump::code_identifier()`, which was
previously stubbed out for Linux dumps. This change keeps the debug ID
in the `dump_syms` output the same to match.
R=mark@chromium.org, thestig@chromium.org
BUG=
Review URL: https://codereview.chromium.org/1688743002 .
Based on changes for ARM, ARM64 and X86, the support for
MIPS and MIPS64 is added in microdump.
TEST=microdump_stackwalk ~/microdump-mips32.dmp symbols/
BUG=microdump_stackwalk failing for mips architectures
Review URL: https://codereview.chromium.org/1731923002/
Some of the symbols in the stack trace are not found in the .dynsym
section but were located in the full symbol table .symtab section
instead. This was causing some of our stack traces to be incomplete or
point to incorrect function names.
Since we only output function names, there are actually not that many
more symbols located in .symtab that aren't in .dynsym. It is better to
simply output all symbols found so our stack traces are complete.
R=mark@chromium.org, thestig@chromium.org
BUG=561447
Review URL: https://codereview.chromium.org/1824063002 .
Patch from David Yen <dyen@chromium.org>.
The code as it stands allocates a chunk of memory of arbitrary size and places an object into it. It stores a pointer to that object and memory into a list telling the compiler that it is a pointer to a char. When the compiler deletes the objects in the list it thinks that the list contains pointers to chars - not pointers to arbitrarily sized regions of memory.
This is fixing an issue that will reproduces when the following optimization (C++ sized dealocation) is enabled: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2013/n3536.html
The fix is to explicitly call the non-sized delete operator, and the library code that supports malloc/free/new/delete will figure out the size of the block of memory from the pointer being passed in.
Patch provided by Darryl Gove.
R=mark@chromium.org
Review URL: https://codereview.chromium.org/1788473002 .
The Linux dumpers use absolute paths for shared libraries referenced by
dumps, so they fail to locate them if the crash originated in a chroot.
This CL enables callers to specify a root prefix, which is prepended to
mapping paths before opening them.
BUG=chromium:591792
TEST=make check
Review URL: https://codereview.chromium.org/1761023002/
Properly handle microdump processing, when the system_log file contains an incomplete microdump section at the top. The processor will process the first complete microdump section.
R=primiano@chromium.org
Review URL: https://codereview.chromium.org/1742843002 .
Because tools/windows/symupload/symupload.cc uses `nullptr` (which
requires VS2010), the CLSID comparison is only performed for msdia100.dll
and later. When compiling with an older (or future) CLSID_DiaSource, we
retain the existing behaviour (i.e. fail if CoCreateInstance fails).
R=ivanpe@chromium.org
BUG=https://bugzilla.mozilla.org/show_bug.cgi?id=1236343
Linux make check is failing for mips, mips64, arm, arm64
with error:
"fatal error: mach/arm/vm_types.h: No such file or directory" in case of arm,
"../src/third_party/mac_headers/mach/machine/vm_types.h:37:2: error: #error architecture not supported" in case of mips/mips64
This was partially fixed in https://codereview.chromium.org/1645673002/.
Here excluding src/common/mac/macho_reader_unittest for hosts other than x86/x86-64.
BUG=make check failure for linux mips
TEST=make check pass
Review URL: https://codereview.chromium.org/1692933002 .
G GL_VERSION|GL_VENDOR|GL_RENDERER.
The GPU version, vendor and renderer are extracted during microdump parsing and populated in the appropriate fields in the SystemInfo struct.
This is to match the changes introduced in crrev.com/1343713002 and crrev.com/1334473003
BUG=chromium:536769
R=primiano@chromium.org
Review URL: https://codereview.chromium.org/1678463002 .
This patch changes MDCVInfoELF (which is currently unused, apparently
a vestigal bit of code landed as part of Solaris support) into a supported
CodeView format that simply contains a build id as raw bytes.
Modern ELF toolchains support build ids nicely:
https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/6/html/Developer_Guide/compiling-build-id.html
It would be useful to have the original build ids of loaded modules in
Linux minidumps, since tools like Fedora's darkserver allow querying by build
id and the current Breakpad code truncates the build id to the size of a GUID,
which loses information:
https://darkserver.fedoraproject.org/
A follow-up patch will change the Linux minidump generation code to produce
MDCVInfoELF in minidumps instead of MDCVInfoPDB70. This patch should be landed
first to ensure that crash processors are able to handle this format before
dumps are generated containing it.
The full build id is exposed as the return value of Minidump::code_identifier(),
which currently just returns "id" for modules in Linux dumps. For
backwards-compatibility, Minidump::debug_identifier() continues to treat
the build id as a GUID, so debug identifiers for existing modules will not
change.
BUG=
R=mark@chromium.org
Review URL: https://codereview.chromium.org/1675413002 .
